Coinbase Engineer's Fly-Brain Model Trades Bitcoin (BTC) With 166,700 Simulated Neurons

166,700 Neurons Read the Candlestick Chart

A software engineer at Coinbase has connected a digital replica of a fruit fly's brain to a complete Bitcoin guide subject's trading account — and the whole experiment is open-source. The project, called Stonkfly, was built by Alex Wormuth, who published the full code on GitHub together with a public dashboard anyone can audit. The simulation is anchored in a connectome: a published wiring map of an adult male fruit fly's nervous system that neuroscientists spent years assembling. Instead of pointing that map at a lab task, Wormuth aimed it at a candlestick chart for Bitcoin news watchers to follow. The model runs 166,700 simulated neurons and roughly 25.6 million connections between them. Stonkfly renders the price chart as a small image and splits it across the fly's two simulated eyes, where light-sensing cells read the raw pixel colors; the brain then settles on buy, sell or hold. Incentives are explicit: a profitable trade fires 15 reward neurons in a burst of virtual dopamine, while a loss activates two neurons that register something unpleasant, with trading fees counted as losses. Notably, the fly cannot borrow funds or bet on falling prices — a constraint set that, on paper, puts it ahead of many AI bots trading equities, which routinely deploy leverage and short positions.

Paper Money, $10 Caps and a Flat Dashboard

Deployment is deliberately cautious. The program runs on paper money by default, beginning with a $100 balance. Live mode caps every order at $10 and restricts attempts to 24 per day, and as of Saturday the public dashboard still showed zero completed trades. Wormuth himself tempers expectations, writing in the repository that “synaptic changes do not establish that it learns to trade profitably,” and acknowledging the experiment does not demonstrate everything. That restraint lands differently in a soft market. Bitcoin (BTC) changed hands near $77,286 on Saturday, up 0.6% on the day but still far below its peak — well under the $79,809 snapshot in Strategy's revised Bitcoin investor guide we covered earlier this month. The AI-trading theme is also drawing broader attention: our desk recently tracked Moonshot AI's police report over Kimi K3 founder rumors as the Bitcoin AI trade drew scrutiny, and a macro overlay looms too, with Goldman Sachs predicting a 25 basis point Fed hike at the September 16 FOMC — conditions where even a conservative $10-per-order bot faces a choppy tape. Readers tracking the market in real time can follow live spot and futures prices on Gate.

Zero Trades, One Open Question

Two threads converge here: a market long defined by HODL patience and Bitcoin maximalism's faith in fixed rules now hosts an experiment asking whether mapped biological circuitry can learn a reward evolution never gave it. Our reading of the primary record — the GitHub repository and its public dashboard — shows hard-coded limits and zero completed trades as of Saturday, so every claim about the fly's skill remains unproven by design. The number to watch next: whether those 24 daily attempts eventually produce a first live fill.